Power prediction credibility-oriented evaluation method, system, equipment and medium

Abstract

The invention relates to the technical field of power systems, and discloses a power prediction credibility-oriented evaluation method, system and device and a medium, and the method comprises the steps: obtaining the meteorological characteristics of a prediction target day in response to a received to-be-evaluated power prediction result; calculating similarities between the meteorological characteristics of the predicted target day and meteorological characteristics corresponding to a plurality of historical days in a historical database to obtain a plurality of meteorological similarities; similar historical days are screened from the multiple historical days based on the corresponding meteorological similarity, a model score corresponding to each similar historical day is obtained, and the model scores are model historical performance scores obtained based on prediction accuracy quantification of the power prediction model on the corresponding similar historical days; calculating the target credibility according to the meteorological characteristics of the similar historical days and the corresponding model scores; calculating a penalty factor according to the model score of each similar historical day; and calculating a credibility score of the to-be-evaluated power prediction result according to the target credibility and the penalty factor, thereby realizing effective evaluation of the power prediction credibility.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and specifically to an assessment method, system, device, and medium for power prediction reliability. Background Technology

[0002] In power systems, particularly in power forecasting for new energy sources (such as photovoltaics and wind power), the current mainstream approach involves using machine learning / artificial intelligence (AI) technology to learn the mapping relationship between weather forecast data and actual power, thereby generating power forecast results for the forecast day. Weather similarity, as a model evaluation criterion, is often used to assess the suitability of these machine learning / AI models (e.g., selecting similar samples for model training, evaluating the model's applicability in specific weather scenarios). In addition, the traditional "similar day" forecasting scheme remains widely used. Its core logic lies in calculating the similarity between the forecast day and the weather characteristics (such as irradiance, wind speed, temperature, humidity, etc.) of each day in the historical database, selecting one or more historical days with the most similar weather conditions, and then generating the power forecast value for the forecast day based on the historical actual power data or historical forecast results of these similar days through weighted averaging or other methods. In practical applications, the calculated meteorological similarity is often used as a basis for weighting in the forecasting process, and is also simply regarded as a rough reference indicator of the reliability of the forecast results. The direct application scenario of this reliability reference is to support power production decision-makers (such as dispatchers, operation and maintenance personnel, and managers) in "selecting the reported results", that is, determining which model's forecast results should be selected for reporting. It is necessary to further judge whether the forecast results are credible, and make corresponding decisions when they are credible, thereby strengthening risk management and ensuring the safe and stable operation of power production.

[0003] However, existing technologies have significant limitations in using meteorological similarity for credibility assessment. For example, the assessment dimensions are relatively singular, relying excessively on meteorological matching degree while failing to incorporate the crucial factor of the power prediction model's historical performance on similar days. This means that even with highly similar meteorological conditions, if the model's actual predictive performance on a historical day is poor, the credibility of the prediction results remains questionable, and the assessment conclusions may be distorted. Furthermore, when assessing based on multiple similar days, existing methods typically only focus on the average level of the model's historical performance, failing to quantify its volatility and stability across different dates. Consequently, it is difficult to reveal the potential risks inherent in model performance instability.

[0004] In summary, existing technologies fail to effectively integrate multiple dimensions such as meteorological matching degree, model historical performance, and stability, making it impossible to output a unified, quantitative, and comprehensive reliability score. This hinders decision-makers from quickly and intuitively grasping the overall reliability of forecast results, thus affecting the accuracy and efficiency of power dispatching and trading decisions. Therefore, there is an urgent need for a power forecast reliability assessment scheme that can integrate multi-dimensional information, quantitatively assess fluctuation risks, and provide intuitive decision support. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for evaluating the reliability of power forecasts, in order to address the problem that existing technologies rely solely on single meteorological similarity assessments, which have many shortcomings, making it difficult to achieve a comprehensive quantitative assessment of the reliability of power forecasts and thus seriously affecting power decision-making.

[0006] In a first aspect, the present invention provides a method for evaluating the reliability of power prediction, the method comprising: In response to the received power prediction results to be evaluated, the meteorological characteristics of the target date are obtained, and the target date is associated with the power prediction results to be evaluated; The similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database is calculated to obtain multiple meteorological similarity scores. Similar historical days are selected from multiple historical days based on corresponding meteorological similarity, and the model score corresponding to each similar historical day is obtained. The model score is the historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days. The reliability of the target is calculated based on the meteorological characteristics of each similar historical day and the corresponding model score; The penalty factor is calculated based on the model scores of similar historical days. The penalty factor is negatively correlated with the stability of the power prediction model. The credibility score of the power prediction results to be evaluated is calculated based on the target credibility and the penalty factor.

[0007] This invention obtains meteorological characteristics of the target date associated with the power prediction result to be evaluated, and calculates the meteorological similarity between these characteristics and the meteorological characteristics of multiple historical days in the historical database. This allows for direct focus on key influencing factors in power prediction, avoiding generalized evaluations detached from practical application scenarios and ensuring a high degree of consistency between the evaluation results and actual prediction credibility. By filtering similar historical days based on meteorological similarity, the model score obtained from the historical prediction accuracy quantification under such dates can be reused, further reducing errors caused by single-dimensional evaluation or subjective judgment. Simultaneously, the comprehensive calculation based on the meteorological characteristics of multiple similar historical days and the corresponding model scores further improves the statistical significance and evaluation accuracy of the target credibility. Furthermore, an additional penalty factor is introduced to correct the target credibility, resulting in a credibility score for the power prediction result to be evaluated. This achieves a dual-dimensional evaluation of accuracy and stability, making the final credibility score more comprehensively reflect the credibility of the prediction result, greatly improving evaluation efficiency and providing effective support for the subsequent practical application of power prediction.

[0008] In one optional implementation, the target confidence level is calculated based on the meteorological characteristics of each similar historical day and the corresponding model score, including: For each similar historical day, the meteorological characteristics of similar historical days are used as weighting coefficients to perform weighted fusion processing on the corresponding model scores to obtain the target credibility.

[0009] This invention uses meteorological characteristics of similar historical days as weighting coefficients, which precisely matches the core principle that power prediction accuracy is strongly correlated with meteorological conditions. This type of weighting allocation does not require subjective setting and can ensure that the weighting logic is highly matched with the actual power prediction application scenario. This makes the target credibility more consistent with the real situation of the current prediction scenario, greatly ensuring that the credibility score of the final output can truly and comprehensively reflect the credibility of the power prediction result to be evaluated, and providing effective support for subsequent decision-making.

[0010] In one alternative implementation, a penalty factor is calculated based on the model scores for each similar historical day, including: Calculate the standard deviation of the model scores for all similar historical days; The standard deviation is input into a preset function for calculation, and the penalty factor is output. The preset function includes at least one of linear and exponential functions.

[0011] This invention uses the standard deviation of model scores for all similar historical days as the core input to directly quantify the fluctuation of the model's predictive performance under similar weather scenarios. This effectively makes up for the limitations of evaluating credibility solely through average accuracy, allowing the penalty factor to accurately reflect the robustness of the model and providing key stability dimensions to support the final score.

[0012] In one alternative implementation, a penalty factor is calculated based on the model scores for each similar historical day, including: Calculate the mean and standard deviation of the model scores for all similar historical days; The stability index is obtained by calculating the ratio of the standard deviation to the mean. The stability index is input into a preset function for calculation, and the output is a penalty factor. The preset function includes at least one of linear and exponential functions.

[0013] This invention calculates the stability index by the ratio of standard deviation to mean, which can realize the relative quantification of model stability and effectively avoid the limitations of simply using standard deviation.

[0014] In one alternative implementation, similar historical days are selected from multiple historical days based on corresponding meteorological similarity, including: The meteorological similarity of each historical day is sorted by numerical value to obtain the sorting results; Based on the sorting results, the historical days with the highest similarity are selected as similar historical days.

[0015] This invention sorts the historical days by meteorological similarity values ​​and then selects the preset number of historical days with the highest similarity. This ensures that the selected similar historical days are highly consistent with the meteorological characteristics of the target prediction day, thereby guaranteeing the quality of the input data for subsequent reliability calculations from the source.

[0016] In one optional implementation, a credibility score for the power prediction result to be evaluated is calculated based on the target credibility and the penalty factor, including: Multiply the target confidence level by the penalty factor to obtain the confidence score of the power prediction result to be evaluated.

[0017] The target credibility focusing model of this invention shows a negative correlation between historical accuracy, penalty factor and model stability under similar meteorological scenarios. By binding the target credibility and penalty factor through multiplication, a positive correlation can be achieved between higher accuracy, stronger stability and higher credibility score. This effectively avoids the one-sidedness of single-dimensional evaluation and enables the final credibility score to comprehensively reflect the reliability of the prediction results.

[0018] In one optional implementation, after calculating the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor, the evaluation method for power prediction credibility further includes: The credibility score is output to the power production and operation system or decision support terminal, so that the relevant decision-makers can judge the reliability of the power prediction results to be evaluated based on the credibility score and make corresponding decisions based on the judgment results.

[0019] The credibility score output by this invention can intuitively reflect the reliability of the prediction results, thus enabling decision-makers to quickly determine the outcome without relying on subjective judgment or complex data analysis. This significantly reduces the decision-making cost and further improves decision-making efficiency and accuracy.

[0020] Secondly, the present invention provides an evaluation system for power prediction reliability, the system comprising: The feature acquisition module is used to acquire the meteorological characteristics of the target prediction day in response to the received power prediction result to be evaluated, and the target prediction day is associated with the power prediction result to be evaluated. The similarity calculation module is used to calculate the similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database, and obtain multiple meteorological similarities. The scoring acquisition module is used to filter similar historical days from multiple historical days based on corresponding meteorological similarity, and obtain the model score corresponding to each similar historical day. The model score is the historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days. The credibility calculation module is used to calculate the credibility of the target based on the meteorological characteristics of each similar historical day and the corresponding model score; The penalty factor calculation module is used to calculate the penalty factor based on the model scores of each similar historical day. The penalty factor is negatively correlated with the stability of the power prediction model. The credibility assessment module is used to calculate the credibility score of the power prediction results to be evaluated based on the target credibility and the penalty factor.

[0021] The power prediction reliability assessment system of this invention obtains the meteorological characteristics of the target date associated with the power prediction result to be evaluated, and calculates the meteorological similarity between the meteorological characteristics of the target date and the meteorological characteristics of the target date in the historical database. This allows for direct focus on key influencing factors of power prediction, avoiding generalized assessments detached from actual application scenarios and ensuring a high degree of consistency between the assessment results and the actual prediction reliability. By screening similar historical dates based on meteorological similarity, the model score obtained by quantifying the historical prediction accuracy of the model under such dates can be reused, further reducing errors caused by single-dimensional assessment or subjective judgment. At the same time, the comprehensive calculation based on the meteorological characteristics of multiple similar historical days and the corresponding model scores further improves the statistical significance and assessment accuracy of the target reliability. Furthermore, an additional penalty factor is introduced to correct the target reliability, resulting in a reliability score for the power prediction result to be evaluated. This achieves a two-dimensional assessment of accuracy and stability, making the final reliability score more comprehensively reflect the reliability of the prediction result, greatly improving assessment efficiency and providing effective support for the subsequent practical application of power prediction.

[0022] Thirdly, the present invention provides an electronic device, which includes a controller, the controller including a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform an evaluation method for power prediction reliability as described in the first aspect or any corresponding embodiment.

[0023] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform an evaluation method for power prediction reliability as described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the power prediction reliability evaluation method according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating another method for evaluating the reliability of power prediction according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating a power prediction reliability assessment scheme based on multi-dimensional feature fusion. Figure 4 This is a structural block diagram of a power prediction reliability evaluation system according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This invention provides an embodiment of a method for evaluating the reliability of power prediction. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] This embodiment provides a method for evaluating the reliability of power prediction. Figure 1 This is a flowchart illustrating the power prediction reliability evaluation method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps: Step S101: In response to the received power prediction result to be evaluated, obtain the meteorological characteristics of the prediction target day, and associate the prediction target day with the power prediction result to be evaluated.

[0029] It should be noted that the power prediction result to be evaluated in this embodiment is power prediction data specifically generated for the target prediction date (which is obtained through the output of the power prediction model). That is, the core descriptive object of the power prediction result (such as time, subject) is completely bound to the target prediction date. For example, if the power prediction result to be evaluated is "24-hour power prediction curve of a photovoltaic power station on October 15, 2024", then the target prediction date is October 15. In addition, the prediction subject of the prediction result to be evaluated in this embodiment (such as power station, regional power grid) is consistent with the subject to which the meteorological characteristics of the target prediction date belong. That is, the meteorological characteristics of the target prediction date (such as irradiance and wind speed of a certain area on October 15) must be the meteorological data of the power station / region corresponding to the prediction result to be evaluated. For example, if the prediction result to be evaluated is "power prediction of a certain power plant on October 15, 2024", then the meteorological characteristics of the target prediction date (October 15) must be the meteorological data of "the area where the power plant is located" on October 15 (while meteorological data of other areas). Note that the specific content of the meteorological features in this embodiment can be adapted to actual needs. For example, meteorological features may include irradiance, wind speed, temperature, humidity, etc., and are only used as examples.

[0030] Step S102: Calculate the similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database to obtain multiple meteorological similarities.

[0031] In this embodiment, the calculation method of similarity can be adaptively determined according to actual requirements. For example, the Euclidean distance can be calculated (that is, to quantify the spatial distance between the "predicted target day meteorological feature vector" and the "historical day meteorological feature vector", and the smaller the distance, the higher the similarity), the cosine similarity (used to measure the "direction consistency" of two feature vectors, and the similarity is quantified by calculating the cosine value of the vector angle. Among them, the smaller the angle, the closer the cosine value is to 1, indicating that the change trends / patterns of meteorological features are more similar), the Pearson correlation coefficient (measures the "linear correlation degree" of two feature vectors, and is suitable for power prediction scenarios where there is a "linear association trend" in meteorological features, such as the positive correlation between irradiance and temperature, and the negative correlation between wind speed and humidity for pattern matching), etc., which are only for illustrative purposes.

[0032] Step S103: Screen similar historical days from multiple historical days based on the corresponding meteorological similarity, and obtain the model score corresponding to each similar historical day. The model score is a model historical performance score quantified based on the prediction accuracy of the power prediction model on the corresponding similar historical day.

[0033] In this embodiment, the acquisition logic of the model score is as follows: First, calculate the prediction accuracy through the "predicted value and true value" of the historical day, and then standardize the accuracy into a score of a unified magnitude (such as the 0-1 interval), and finally bind and store it with the historical day and the power prediction model for direct call when screening similar historical days later. Among them, the prediction accuracy can be calculated through common model evaluation indicators, such as the mean absolute percentage error (the most commonly used error indicator in the power industry), the root mean square error (focusing on the "absolute deviation size" of power prediction), the hit rate (only focusing on "whether the predicted value is within the allowable deviation range", such as when the power grid requires the power prediction deviation ≤ 5% to be qualified, which has the significant advantages of simple logic and easy understanding), etc. (For example, if the mean absolute percentage error of the hourly power prediction of a certain historical day is 8%, then the accuracy is 1 0.08 = 0.92). In addition, to standardize the prediction accuracy into the model score, the accuracy can be directly used as the model score, such as the accuracy and the model score are both 0.92; or threshold optimization can be performed to filter extreme outliers. For example, if there are extreme meteorological conditions on a historical day, such as typhoons, blizzards, etc., resulting in a very low prediction accuracy (accuracy < 0.3), a minimum score threshold (such as 0.1) can be set to avoid the influence of too low scores on subsequent weighted calculations; at the same time, a maximum threshold (such as 1) is set, then the model score = max(minimum score threshold, min(accuracy, maximum threshold)), such as the model score = max(0.1, min(0.3, 1)) = 0.3, which is only for illustrative purposes.

[0034] Step S104: Calculate the target credibility according to the meteorological features and the corresponding model scores of each similar historical day.

[0035] It should be noted that the essence of the target credibility in this embodiment is based on "historical days with similar meteorological characteristics to the predicted target day". The credibility of the power prediction result to be evaluated is quantified in terms of accuracy by weighted fusion of the corresponding model scores through meteorological feature similarity. The specific weighted fusion method can be adaptively adjusted according to actual needs.

[0036] Step S105: Calculate the penalty factor based on the model scores of each similar historical day. The penalty factor is negatively correlated with the stability of the power prediction model.

[0037] It should be noted that the penalty factor in this embodiment is a quantitative correction term for the stability of the power prediction model. It aims to measure the stability of the model's prediction performance under similar weather scenarios by observing the fluctuation of the model score on similar historical days, and then correct the target credibility (i.e., the base score of the accuracy dimension). The worse the stability of the power prediction model, the smaller the penalty factor, and the stronger the weakening of the target credibility; conversely, the better the stability, the closer the penalty factor is to 1, and the smaller the impact on the target credibility.

[0038] Step S106: Calculate the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor.

[0039] The power prediction reliability assessment method of this invention obtains the meteorological characteristics of the target date associated with the power prediction result to be evaluated, and calculates the meteorological similarity between the meteorological characteristics of the target date and the meteorological characteristics of the target date in the historical database. This allows for direct focus on key influencing factors of power prediction, avoiding generalized assessments detached from actual application scenarios, and ensuring that the assessment results are highly consistent with the actual prediction reliability. By screening similar historical dates based on meteorological similarity, the model score obtained by quantifying the historical prediction accuracy of the model under such dates can be reused, further reducing errors caused by single-dimensional assessment or subjective judgment. At the same time, the comprehensive calculation based on the meteorological characteristics of multiple similar historical days and the corresponding model scores further improves the statistical significance and assessment accuracy of the target reliability. Furthermore, an additional penalty factor is introduced to correct the target reliability, resulting in a reliability score for the power prediction result to be evaluated. This achieves a two-dimensional assessment of accuracy and stability, making the final reliability score more comprehensively reflect the reliability of the prediction result, greatly improving assessment efficiency, and providing effective support for the subsequent practical application of power prediction.

[0040] This embodiment provides a method for evaluating the reliability of power prediction. Figure 2 This is a flowchart illustrating another method for evaluating the reliability of power prediction according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps: Step S201: In response to the received power prediction result to be evaluated, obtain the meteorological characteristics of the target prediction date, and associate the target prediction date with the power prediction result to be evaluated. For details, please refer to [link to details]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0041] Step S202 involves calculating the similarity between the meteorological characteristics of the predicted target day and the corresponding meteorological characteristics of multiple historical days in the historical database, resulting in multiple meteorological similarity scores. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0042] Step S203: Select similar historical days from multiple historical days based on corresponding meteorological similarity, and obtain the model score corresponding to each similar historical day. The model score is the historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days.

[0043] In this embodiment, step S203 above, which involves filtering similar historical days from multiple historical days based on corresponding meteorological similarity, includes: Step a1: Sort the meteorological similarity of each historical day according to the numerical value to obtain the sorting result.

[0044] In this embodiment, the meteorological similarity values ​​can be sorted from largest to smallest or from smallest to largest, and the specific sorting method is adaptively set based on actual needs.

[0045] Step a2: Based on the sorting results, select the historical days with the highest similarity as similar historical days.

[0046] It should be noted that the specific value of the preset quantity in this embodiment can be adjusted adaptively according to actual needs. For example, the meteorological similarity of each historical day can be sorted from largest to smallest, and the historical days corresponding to the top 5 meteorological similarity values ​​can be selected as similar historical days.

[0047] In this embodiment of the invention, by sorting the historical days by meteorological similarity values ​​and then selecting a preset number of historical days with the highest similarity, the selected similar historical days can be ensured to the greatest extent that their meteorological characteristics are highly consistent with the meteorological characteristics of the target prediction day, thereby guaranteeing the quality of the input data for subsequent reliability calculations from the source.

[0048] Step S204: Calculate the target credibility based on the meteorological characteristics of each similar historical day and the corresponding model score.

[0049] Specifically, step S204 includes: Step S2041: For each similar historical day, the meteorological characteristics of the similar historical day are used as weighting coefficients to perform weighted fusion processing on the corresponding model score to obtain the target credibility.

[0050] It should be noted that the weighted fusion processing steps in this embodiment include: since the meteorological feature similarity is the original similarity value (which may not sum to 1), it needs to be standardized to "normalized weights" first to ensure that the weight ratio of each similar historical day is reasonable; then, the "normalized weight" of each similar historical day is multiplied by the "model score", and all product results are summed to obtain the target credibility, where the target credibility value ranges from 0 to 1. The closer it is to 1, the higher the accuracy of the current prediction result based on historical similar scenarios, and the stronger the credibility of the corresponding result. Note that the above weighted fusion processing method is only an example and can be adjusted adaptively according to actual needs.

[0051] In this embodiment, meteorological characteristics of similar historical days are used as weighting coefficients, precisely aligning with the core principle that "power prediction accuracy is strongly correlated with meteorological conditions." Meteorological characteristics are not only the core dimension for selecting similar historical days but also a key objective factor influencing model prediction performance. This type of weight allocation requires no subjective setting and is entirely based on the essential influencing factors of the scenario, ensuring a high degree of match between the weighting logic and the actual application scenario of power prediction, making the target credibility more closely reflect the real situation of the current prediction scenario. Simultaneously, "differentiated fusion" is achieved through meteorological characteristic weighting, meaning that similar historical days with meteorological characteristics closer to the target day have a higher weighting in the model score (because the model performance on such historical days has greater reference value for the current prediction). Compared to the traditional method of "simply averaging the scores of all similar days," this effectively avoids the dilution of the accuracy of the evaluation results by low-correlation historical data, making the target credibility more reflective of the model's true capabilities in the current prediction scenario and significantly improving the accuracy of the evaluation results.

[0052] Step S205: Calculate the penalty factor based on the model scores of each similar historical day. The penalty factor is negatively correlated with the stability of the power prediction model.

[0053] It should be noted that the penalty factor is calculated in two ways in this embodiment. When the average score of the model is stable on similar historical days (e.g., the mean is concentrated in the range of 0.8 to 0.9), the magnitude is consistent (no significant difference), the computational efficiency requirement is high, and there is no need to compare different mean scenarios horizontally (e.g., short-term similar scenario evaluation of the same model), the stability index (i.e., standard deviation / mean, also known as the coefficient of variation) is not calculated. This type of calculation aims to quickly quantify the absolute stability of the model score, simplify the calculation process in the case of stable mean, take into account the evaluation efficiency and the basic stability correction requirements, and avoid the redundancy caused by overly complex calculations. Furthermore, for scenarios where the average scores of similar historical days differ significantly (e.g., the average scores of different time periods / models fluctuate between 0.7 and 0.95), where it is necessary to compare the stability under different means, or where the magnitude of model scores is inconsistent (e.g., some historical day scores are concentrated between 0.6 and 0.7, while others are between 0.85 and 0.95), the relative stability of the model under different score levels is accurately quantified through calculation with the coefficient of variation. This ensures that the stability assessment is not affected by the difference in means, and provides a more realistic basis for stability correction to the target credibility.

[0054] In this embodiment, when the coefficient of variation is not calculated, the penalty factor calculated based on the model score for each similar historical day in step S205 above includes: Step b1: Calculate the standard deviation of the model scores for all similar historical days.

[0055] In this embodiment, the standard deviation is calculated by referring to conventional calculation methods in the art for suitability determination.

[0056] Step b2: Input the standard deviation into a preset function for calculation and output the penalty factor. The preset function includes at least one of linear and exponential functions.

[0057] In this embodiment of the invention, the standard deviation of the model scores for all similar historical days is used as the core input to directly quantify the fluctuation of the model’s prediction performance under similar weather scenarios. This effectively makes up for the limitations of evaluating credibility solely through average accuracy, allowing the penalty factor to accurately reflect the robustness of the model and provide key stability dimension support for the final score.

[0058] In this embodiment, when calculating the coefficient of variation, step S205 above, which calculates the penalty factor based on the model score for each similar historical day, includes: Step c1: Calculate the mean and standard deviation of the model scores for all similar historical days.

[0059] Step c2: Calculate the ratio of the standard deviation to the mean to obtain the stability index.

[0060] Step c3: Input the stability index into a preset function for calculation and output a penalty factor. The preset function includes at least one of linear functions and exponential functions.

[0061] In this embodiment of the invention, the stability index is obtained by calculating the ratio of the standard deviation to the mean, which can more objectively reflect the relative stability of the model in similar scenarios. This makes the calculation of the penalty factor more consistent with the actual performance of the model and effectively avoids misjudgment of stability caused by the difference in mean.

[0062] Step S206: Calculate the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor.

[0063] Specifically, step S206 includes: Step S2061: Multiply the target confidence level by the penalty factor to obtain the confidence score of the power prediction result to be evaluated.

[0064] In this embodiment of the invention, the historical accuracy, penalty factor and model stability of the target credibility focusing model under similar meteorological scenarios are negatively correlated. By binding the target credibility and penalty factor through multiplication, a positive correlation can be achieved between higher accuracy, stronger stability and higher credibility score. This effectively avoids the one-sidedness of single-dimensional evaluation and enables the final credibility score to comprehensively reflect the reliability of the prediction results.

[0065] In this embodiment, the ultimate goal of obtaining power prediction results is to provide support for "judging whether the prediction results are reliable." It is necessary to output the credibility score so that power production decision-makers (such as dispatchers, maintenance personnel, and managers) can quickly determine whether the prediction results are credible and make corresponding decisions, thereby strengthening risk control and ensuring the safe and stable operation of power production. Therefore, after calculating the credibility score of the power prediction results to be evaluated based on the target credibility and penalty factor, the power prediction credibility evaluation method in this embodiment further includes: outputting the credibility score to the power production operation system or decision support terminal, so that the relevant decision-makers can judge the reliability of the power prediction results to be evaluated based on the credibility score and make corresponding decisions based on the judgment results. Specifically, this embodiment outputs a credibility score to intuitively reflect the reliability of the prediction results, thereby eliminating the need for decision-makers to rely on subjective judgment or complex data analysis. They can quickly determine the reliability using only the score, significantly reducing the judgment cost for decision-makers and further improving decision-making efficiency and accuracy.

[0066] In one specific embodiment, considering the following main drawbacks of existing quantitative evaluation schemes for the reliability or credibility of power prediction results: (1) The evaluation dimension is too singular and does not consider the historical performance of the model: When evaluating the reliability of the prediction results, the existing schemes rely too much on "meteorological similarity (S)".i This overlooks a crucial issue: even with perfectly matched meteorological conditions (S...),... i (High accuracy), but if the predictive model performs poorly on that historical day (i.e., historical accuracy), then the predictions based on that day are equally unreliable. Therefore, the relevant techniques fail to incorporate the model's historical performance (P... i ) will be included in the evaluation.

[0067] (2) Lack of stability considerations and absence of risk assessment: When using multiple similar days (e.g., K=3 days) for comprehensive evaluation, related techniques (e.g., simple averaging or weighted averaging) only focus on average performance and fail to consider the "stability" of the model's performance across these similar days. For example, one model has historical accuracy rates of 95%, 94%, and 96% on 3 similar days (stable performance), while another model has rates of 99%, 70%, and 90% (unstable performance); their average values ​​may be similar, but the latter's predictive risk is much higher than the former. Therefore, related techniques cannot quantify the risk caused by the volatility of model performance.

[0068] (3) Decision-making is not intuitive: There is a lack of a mechanism to measure meteorological similarity (S) i ), historical performance of the model (P) i The single, quantitative comprehensive credibility score, which organically integrates multiple key dimensions such as model stability (CV), makes it difficult for power dispatchers or traders to make quick, comprehensive, and accurate risk assessments and decisions.

[0069] In summary, to address the aforementioned shortcomings—namely, how to overcome the one-sidedness of relying solely on meteorological similarity for evaluation in related technologies—this embodiment provides a method that comprehensively considers meteorological similarity (S... i ), historical performance of the model (P) i This paper presents a quantitative evaluation scheme for power prediction reliability, including the model's multi-day performance stability (CV), to output a more accurate, secure, and intuitive single decision indicator. It should be noted that this technical solution creatively proposes to consider not only meteorological similarity (S) when calculating reliability. i Furthermore, it is necessary to introduce the model's "historical performance score" (P) on that historical day. i "); At the same time, when using multi-day data, it is necessary to quantify the model's historical performance on multiple similar days (P)"; i The stability of meteorological similarity (S) is assessed, and the coefficient of variation (CVp) is used as a measure of this stability. Finally, a multi-dimensional feature fusion method is employed, namely: firstly, meteorological similarity (S) is used as the metric; secondly, the stability of meteorological similarity (S) is assessed. i ) are used as weights to account for the model's historical performance (P) iThe weighted average is used to obtain the "BaseConf"; then, the "Stability Penalty Factor (Factor_stab)" is calculated based on the "Coefficient of Variation (CVp)"; finally, the two are multiplied together (FinalConf = BaseConf × Factor_stab) to obtain the final comprehensive confidence. Figure 3 This is a flowchart illustrating a power prediction reliability assessment scheme based on multi-dimensional feature fusion. As shown in the diagram, the process includes: S100: Data Preparation. In this embodiment, this step first obtains the target meteorological feature vector V for the day to be predicted. target Specifically, meteorological feature vectors V for N historical days i (i=1…N) are obtained from the historical database. i And N corresponding historical performance scores P generated by the prediction model M to be evaluated on historical day i. i Note that this P i It is a quantification of the model's historical accuracy, such as P. i =1-MAPE i (Where, MAPE is the mean absolute percentage error).

[0070] S200: Similarity calculation. Traverse the historical database and calculate the target meteorological vector V. target With each historical meteorological vector V i Meteorological similarity S between i Note that similarity can be calculated using techniques known in the field, such as cosine similarity or the reciprocal of the normalized Euclidean distance.

[0071] S300: Similarity Day Filtering. Based on the calculated similarity S i We select the K most similar historical days in descending order (where K is a preset integer, such as K=3 or 5); thus obtaining a similarity set {s1, s2, …, s}. K} and a corresponding set of historical performance ratings {p1, p2, …, p K}

[0072] S400: Calculates the BaseConfidence. It fuses data from K similar days to calculate a comprehensive meteorological similarity (S). K ) and historical performance of the model (p K The basic credibility of the weather similarity (s). For example, this embodiment uses a weighted average method, where the weather similarity s K This is used as a weight to reflect the principle that "the more similar the dates, the greater the influence." Specifically, s is used. K The square of s is used as the numerator weight. KAs the weight in the denominator to enhance the influence of days with high similarity, the formula for calculating the base confidence (BaseConf) is as follows: .

[0073] It should be noted that meteorological similarity s can also be used in this embodiment. K As a unified weight for the numerator and denominator, the formula for calculating the BaseConf confidence is: .

[0074] S500: Calculate the coefficient of variation (CVp). In this embodiment, to quantify the volatility risk of the model over K similar days, this step calculates K historical performance scores {p1, p2, …, p}. K The coefficient of variation (CVp) of}. This step should follow the scientific terminology and calculation methods uniformly adopted in the relevant technical field. The calculation process is as follows: S510: Calculate the arithmetic mean μp of K historical performance scores.

[0075] S520: Calculate the standard deviation σp of K historical performance scores.

[0076] S530: Calculate the coefficient of variation CVp = σp / μp.

[0077] S600: Calculate the penalty factor Factor_stab. In this embodiment, a penalty factor between 0 and 1 is calculated based on the coefficient of variation (CVp). Note that this factor is negatively correlated with CVp; that is, the greater the volatility (the larger the CVp), the stronger the penalty (the smaller the factor). For example, in this embodiment, a linear penalty formula (i.e., a linear function) can be used for calculation, such as the formula for calculating the penalty factor Factor_stab:

[0078] Wherein, ω is an adjustable penalty weight coefficient (such as ω=1 or ω=1.5) used to adjust the sensitivity to instability.

[0079] It should be noted that, in this embodiment, the index used to quantify stability in step S500, besides the "coefficient of variation (CVp)," can also be replaced by the "standard deviation (σp)." In this case, the calculation formula for the penalty factor in step S600 can be adjusted accordingly:

[0080] Where, ω s This is an adjustable penalty weighting coefficient.

[0081] It should be noted that the penalty factor Factor_stab in this embodiment can also be calculated using a non-linear function, such as an exponential decay function, to provide a smoother penalty curve for the growth of the coefficient of variation CVp; then the calculation formula for the penalty factor can be adjusted accordingly: .

[0082] S700: FinalConfidential Score. This embodiment multiplies the base credibility score by a stability penalty factor to obtain the final comprehensive credibility score that integrates all dimensions. The specific calculation formula is as follows: .

[0083] S800: Output Results. In this embodiment, the calculated FinalConf (usually a percentage, such as 77.4%) is output to the power dispatching system, trading system, or user interface as the core basis for decision-makers to judge the reliability of this prediction result.

[0084] In summary, the quantitative evaluation scheme for the reliability or credibility of power prediction results in this invention has the following advantages: (1) More comprehensive evaluation dimensions and more accurate results: By introducing the "model historical performance score (P)", i This addresses the issue that related technologies rely solely on "meteorological similarity (S)". i This avoids the one-sidedness of the problem of "the weather is very similar, but the model has a poor historical performance", and makes the assessment results closer to the real risk.

[0085] (2) Effectively quantifies the stability risk of the model: By creatively introducing the "coefficient of variation (CVp)" and the "stability penalty factor (Factor_stab)," the problem that related techniques cannot assess the volatility of model performance is solved; for example, if a model's performance fluctuates over multiple days (e.g., P), the model's stability risk is effectively quantified. i (If the percentages are 99% and 70% respectively), their CVP will be very high, resulting in a very low Factor_stab, which will significantly lower the final credibility of FinalConf, thus issuing a clear risk warning to the user.

[0086] (3) More intuitive and efficient decision support: Through the final output, a decision support system integrating S i (Weight), P i The single, quantitative comprehensive credibility score (FinalConf) of the base score and CVP (penalty) allows power dispatchers or traders to clearly judge the reliability of the prediction results based on the score, which greatly improves decision-making efficiency and security.

[0087] This embodiment also provides an evaluation system for power prediction reliability, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, a "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0088] This invention provides an evaluation system for power prediction reliability, such as... Figure 4 As shown, the system includes: The feature acquisition module 401 is used to acquire the meteorological characteristics of the target prediction day in response to the received power prediction result to be evaluated, and the target prediction day is associated with the power prediction result to be evaluated.

[0089] The similarity calculation module 402 is used to calculate the similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database, and obtain multiple meteorological similarities.

[0090] The scoring acquisition module 403 is used to filter similar historical days from multiple historical days based on corresponding meteorological similarity, and to obtain the model score corresponding to each similar historical day. The model score is the historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days.

[0091] The credibility calculation module 404 is used to calculate the credibility of the target based on the meteorological characteristics of each similar historical day and the corresponding model score.

[0092] The penalty factor calculation module 405 is used to calculate the penalty factor based on the model scores of each similar historical day. The penalty factor is negatively correlated with the stability of the power prediction model.

[0093] The credibility assessment module 406 is used to calculate the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor.

[0094] In some alternative implementations, the scoring acquisition module 403 includes: The first filtering submodule is used to sort the meteorological similarity of each historical day according to the numerical value, and obtain the sorting results.

[0095] The second filtering submodule is used to filter out the historical days with the highest similarity based on the sorting results, using a preset number of such historical days as similar historical days.

[0096] In some optional implementations, the credibility calculation module 404 includes a credibility calculation submodule, which is used to perform weighted fusion processing on the corresponding model score for each similar historical day, using the meteorological characteristics of similar historical days as weight coefficients, to obtain the target credibility.

[0097] In some optional implementations, the penalty factor calculation module 405 includes: The first operation submodule is used to calculate the standard deviation of the model scores corresponding to all similar historical days.

[0098] The second operation submodule is used to input the standard deviation into a preset function for calculation and output a penalty factor. The preset function includes at least one of linear functions and exponential functions.

[0099] In some optional implementations, the penalty factor calculation module 405 includes: The first calculation submodule is used to calculate the mean and standard deviation of the model scores corresponding to all similar historical days.

[0100] The second calculation submodule is used to calculate the ratio of the standard deviation to the mean, thus obtaining the stability index.

[0101] The third calculation submodule is used to input the stability index into a preset function for calculation and output a penalty factor. The preset function includes at least one of linear functions and exponential functions.

[0102] In some optional implementations, the credibility assessment module 406 includes a credibility assessment submodule for multiplying the target credibility by a penalty factor to obtain a credibility score for the power prediction result to be evaluated.

[0103] In some optional implementations, the system further includes a scoring output module for outputting a credibility score to a power production operation system or decision support terminal, so that the relevant decision-makers can judge the reliability of the power prediction results to be evaluated based on the credibility score and make corresponding decisions based on the judgment results.

[0104] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0105] The power prediction reliability evaluation system in this embodiment is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0106] The power prediction reliability assessment system of this invention obtains the meteorological characteristics of the target date associated with the power prediction result to be evaluated, and calculates the meteorological similarity between the meteorological characteristics of the target date and the meteorological characteristics of the target date in the historical database. This allows for direct focus on key influencing factors of power prediction, avoiding generalized assessments detached from actual application scenarios and ensuring a high degree of consistency between the assessment results and the actual prediction reliability. By screening similar historical dates based on meteorological similarity, the model score obtained by quantifying the historical prediction accuracy of the model under such dates can be reused, further reducing errors caused by single-dimensional assessment or subjective judgment. At the same time, the comprehensive calculation based on the meteorological characteristics of multiple similar historical days and the corresponding model scores further improves the statistical significance and assessment accuracy of the target reliability. Furthermore, an additional penalty factor is introduced to correct the target reliability, resulting in a reliability score for the power prediction result to be evaluated. This achieves a two-dimensional assessment of accuracy and stability, making the final reliability score more comprehensively reflect the reliability of the prediction result, greatly improving assessment efficiency and providing effective support for the subsequent practical application of power prediction.

[0107] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0108] The following is a detailed reference. Figure 5 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device includes a controller, which may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from memory 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0109] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0110] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the in-vehicle voice testing method of the embodiments of the present invention.

[0111] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0112] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor main control chips, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0113] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for evaluating the reliability of power prediction, characterized in that, The method includes: In response to the received power prediction result to be evaluated, the meteorological characteristics of the predicted target day are obtained, and the predicted target day is associated with the power prediction result to be evaluated; The similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database is calculated to obtain multiple meteorological similarities. Similar historical days are selected from multiple historical days based on corresponding meteorological similarity, and the model score corresponding to each similar historical day is obtained. The model score is the historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days. The credibility of the target is calculated based on the meteorological characteristics of each similar historical day and the corresponding model score; A penalty factor is calculated based on the model scores of each of the aforementioned similar historical days, and the penalty factor is negatively correlated with the stability of the power prediction model; The credibility score of the power prediction result to be evaluated is calculated based on the target credibility and the penalty factor.

2. The method for evaluating the reliability of power prediction according to claim 1, characterized in that, The calculation of target credibility based on the meteorological characteristics of each similar historical day and the corresponding model score includes: For each similar historical day, the meteorological characteristics of the similar historical day are used as weighting coefficients to perform weighted fusion processing on the corresponding model scores to obtain the target credibility.

3. The method for evaluating the reliability of power prediction according to claim 1, characterized in that, The calculation of the penalty factor based on the model score of each of the aforementioned similar historical days includes: Calculate the standard deviation of the model scores for all similar historical days; The standard deviation is input into a preset function for calculation, and a penalty factor is output. The preset function includes at least one of a linear function and an exponential function.

4. The method for evaluating the reliability of power prediction according to claim 1, characterized in that, The calculation of the penalty factor based on the model score of each of the aforementioned similar historical days includes: Calculate the mean and standard deviation of the model scores for all similar historical days; The ratio of the standard deviation to the mean is calculated to obtain the stability index; The stability index is input into a preset function for calculation, and a penalty factor is output. The preset function includes at least one of linear functions and exponential functions.

5. The method for evaluating the reliability of power prediction according to claim 1, characterized in that, The process of filtering similar historical days from multiple historical days based on corresponding meteorological similarity includes: The meteorological similarity of each historical day is sorted by numerical value to obtain the sorting results; Based on the sorting results, the historical days with the highest similarity are selected as the similar historical days.

6. The method for evaluating the reliability of power prediction according to claim 1, characterized in that, The calculation of the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor includes: The confidence score of the power prediction result to be evaluated is obtained by multiplying the target confidence score by the penalty factor.

7. The method for evaluating the reliability of power prediction according to any one of claims 1 to 6, characterized in that, After calculating the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor, the method further includes: The credibility score is output to the power production and operation system or decision support terminal so that the relevant decision-makers can judge the reliability of the power prediction result to be evaluated based on the credibility score and make corresponding decisions based on the judgment result.

8. A power prediction reliability evaluation system, characterized in that, The system includes: The feature acquisition module is used to acquire meteorological characteristics of the target prediction day in response to the received power prediction result to be evaluated, wherein the target prediction day is associated with the power prediction result to be evaluated; The similarity calculation module is used to calculate the similarity between the meteorological characteristics of the predicted target day and the meteorological characteristics of multiple historical days in the historical database, and obtain multiple meteorological similarities. The scoring acquisition module is used to filter similar historical days from multiple historical days based on corresponding meteorological similarity, and to obtain the model score corresponding to each similar historical day. The model score is a historical performance score of the model obtained by quantifying the prediction accuracy of the power prediction model on the corresponding similar historical days. The credibility calculation module is used to calculate the credibility of the target based on the meteorological characteristics of each similar historical day and the corresponding model score; The penalty factor calculation module is used to calculate the penalty factor based on the model score of each of the similar historical days, and the penalty factor is negatively correlated with the stability of the power prediction model. The credibility assessment module is used to calculate the credibility score of the power prediction result to be evaluated based on the target credibility and the penalty factor.

9. An electronic device, characterized in that, The electronic device includes a controller, which includes a memory and a processor, the memory and the processor being communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the power prediction reliability assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the power prediction reliability assessment method according to any one of claims 1 to 7.

Images